On 15 Feb 2019, Andreas said:
> I created a patch to make the bypasses for readahead and background IO
> that were added in late 2017 configurable via SysFS switches. Since
> receiving that original patch in my distro's kernel I noticed
> performance degradation and found a few people asking about similar
> symptoms as I noticed online who weren't able to identify the problem.
Hm.
> diff --git a/drivers/md/bcache/request.c b/drivers/md/bcache/request.c
> index 15070412a32e..8028638b348e 100644
> --- a/drivers/md/bcache/request.c
> +++ b/drivers/md/bcache/request.c
> @@ -394,9 +394,13 @@ static bool check_should_bypass(struct cached_dev *dc, struct bio *bio)
> * Flag for bypass if the IO is for read-ahead or background,
> * unless the read-ahead request is for metadata (eg, for gfs2).
> */
> - if (bio->bi_opf & (REQ_RAHEAD|REQ_BACKGROUND) &&
> - !(bio->bi_opf & REQ_PRIO))
> - goto skip;
> + if (!(bio->bi_opf & REQ_PRIO))
> + {
> + if (dc->bypass_readahead_io && (bio->bi_opf & REQ_RAHEAD))
> + goto skip;
> + if (dc->bypass_background_io && (bio->bi_opf & REQ_BACKGROUND))
> + goto skip;
> + }
The thing you based this on is buggy: so, as a result, your patch is
buggy too. You want to apply this atop the patch in
<https://lkml.org/lkml/2019/2/7/77>, I think: without it, metadata I/O
will often not be cached at all. I also suspect this is the cause of
your performance degradation.
FYI, REQ_BACKGROUND is only used for writeback that I can see, so I'm
not sure that bit of the patch does anything at all, given that
as I understand it bypassing relates solely to reads.
--
NULL && (void)
